ABSTRACT
This chapter outlines the basic conditions for employing parameter estimation methods, describes some of the techniques and evaluates their application for breath sound analysis. Breath sounds that have been normalized to their envelope may be analyzed by those parameter estimation methods that require complying with the conditions. Testing the stability of the mean value of the signal amplitude is usually redundant, since most breath sounds signals are transmitted through high-pass filters, which eliminate the direct current component that may be present in the signal. The parametric representation of breath sounds, using autoregressive modeling, may have significant physiological and clinical implications. Breath sounds are nonstationary signals due to the cyclic action of breathing. Moreover, they contain nonstationary added interference such as ambient noise and heart sounds. A special type of adaptive filter was proposed by Arakawa, Ono, and coworkers, who used a nonlinear filter to separate the slowly changing or quasistationary breath sounds from rapidly changing, highly nonstationary crackles.
